Published Release Notes

Optimized performance of embedded decision strategies

The performance of embedded strategies has now been optimized so that these strategies take less time and fewer CPU and memory resources to complete. This enhancement increases the performance of cloud and on-premises deployments.

For more information, see About Strategy rules.

New method of aggregating real-time events

Event Strategy now uses the approximate median for calculating aggregations. Approximate median calculation replaces the existing method because it requires fewer system resources for saving or loading the event strategy state. In addition, you can now calculate aggregated values by using a new Landmark window type. The window type captures all values of specific event properties from the start of the data flow that references the event strategy.

For more information, see Adding aggregations in event strategies.

Extended language support in text analytics

Pega Platform™ text analytics now supports 9 more languages. You can build entity and topic detection models to analyze text in the Japanese, Russian, Turkish, Danish, Norwegian, Swedish, Polish, Croatian, and Czech languages. In addition, Pega Platform now includes a default small talk detection model for each of these languages.

For more information, see Language support for NLP and Out-of-the-box text analytics models.

Historical data extraction for adaptive models

Adaptive models now support the extraction of historical customer responses for offline analysis. Additionally, you can use the historical data to build a model in a machine learning service of your choice.

For more information, see Extract historical responses from adaptive models for offline analysis (8.4).

Improved reliability of Visual Business Director

Improvements to the reliability of Visual Business Director (VBD) eliminate the impact of temporary access loss to the VBD cluster. The improvements include the ability to continue writing data to a VBD dataset while the corresponding server node is unavailable, enhancements to single-case data flows that write data to VBD, and a reduction in the VBD cache size.

For more information, see Eliminate the impact of temporary access loss to the VBD cluster (8.4).

Strategies are now available as a condition type in proposition filters

When designing proposition filters, you can now include strategy outcomes as a condition to be evaluated. Select a strategy and specify the component that you want to use as output to include more advanced decisioning in your proposition filters.

For more information, see Enrich your Proposition Filter rules with conditions based on strategy results.

Default text analytics model for detecting small talk

Enhance your web chat experience by configuring chatbots to detect small talk. Pega Platform™ provides the default model that is a topic detection model, available for English, German, Spanish, Dutch, Portuguese, Italian, and French.

For more information, see Configure your chatbot for small talk.

Integrate your application with external Kafka clusters

Configure your application to use an external Kafka cluster for managing real-time data. By having an external Stream service provider, you can perform such maintenance tasks as an upgrade or hotfix deployment faster because external nodes do not require restarting.

For more information, see Externally managed Stream service.

Updated architecture of the data flow service

Benefit from improvements to data flow architecture that increase the stability of data flow runs and minimize the need for manual restarting of data flow jobs. Real-time data flows now use improved node rebalancing for better handling of failed or restarted nodes. If the topology changes, batch data flows no longer attempt to pause and resume the run. As such, there are fewer interactions with the database and between the nodes, resulting in the increased resilience of the Data Flow service.

If you are upgrading from a previous version of Pega Platform™, see Changes to the architecture of the Data Flow service for an overview of the changes to the Data Flow service compared to previous versions.

Changes to the architecture of the Data Flow service

In Pega Platform™ 8.4, the architecture of batch and real-time data flows uses improved node handling to increase the stability of data flow runs. As a result, there are fewer interactions with the database and between the nodes, resulting in increased resilience of the Data Flow service.

If you upgrade from a previous version of Pega Plaftorm, see the following list for an overview of the changes in the behavior of the Data Flow service compared to previous versions:

• Responsiveness
• Updates to lifecycle actions
• Starting a run
• Triggering pre- and post-activities
• Selecting a node fail policy
• No service nodes and active runs

Responsiveness

Nodes no longer communicate and trigger each other, but run periodic tasks instead. As such, triggering a new run does not cause the service nodes to immediately start the run. Instead, the run starts a few seconds later. The same applies to user actions such as stopping, starting, and updating the run. The system also processes topology changes as periodic tasks, so it might take a few minutes for new nodes to join runs, or for partitions to redistribute when a node leaves a run.

Updates to lifecycle actions

To make lifecycle actions more intuitive, the Stop action consolidates both the Stop and Pause actions. The Start action consolidates both the Resume and Start actions.

You can resume or restart stopped and failed runs with the Start and Restart actions. The Start action is only available for resumable runs and continues the run from where it stopped. The Restart action causes the run to process from the beginning. Completed runs can only be restarted. If a run completes with failures, you can restart it from the beginning, or process only the errors by using the Reprocess failures action.

Starting a run

New data flow runs have the Initializing status, and start automatically. You no longer need to manually start a new run, so the New status is now removed.

If there are no nodes available to process a run, the run gets the Queued status and waits for an available node.

Triggering pre- and post-activities

The system now triggers pre-activities on a random service node, rather than on the node that triggered the run.

The system triggers post-activities only for runs that complete, fail, or complete with failures. If you manually stop a run with the Stop action, the post-activity does not trigger. However, restarting the run with the Restart action triggers first the post-activity, and then the pre-activity.

You can no longer choose to run pre- and post-activities on all nodes.

Selecting a node fail policy

For resumable runs, you can no longer select a node fail policy. If a node fails, the partitions assigned to that node automatically continue the run on different nodes.

For non-resumable runs, you can choose to restart the partitions assigned to the failed node on different nodes, or to fail the partitions assigned to the failed node.

No service nodes and active runs

If the last data flow node for an in-progress run fails, the run remains in the In Progress state, even if no processing takes place. This behavior results from the fact that data flow architecture now prevents unrelated nodes from affecting runs.